Lesson 5.3Lesson 27 of 30
Error propagation across agents
Implement error propagation strategies across multi-agent systems
Key points
- 1A subagent that cannot finish reports structured context: failure type, the query it attempted, partial results and alternative approaches. The coordinator needs these to choose a recovery: retry, rephrase, try another source or proceed with a gap.
- 2An access failure (timeout, 403, source down) is not the same as a valid empty result (the query ran and matched nothing). Report them distinctly, or the coordinator will treat "could not search" as "nothing exists".
- 3Generic statuses like "search unavailable" or
status: errorhide the context the coordinator needs; it can only retry everything blindly or give up. - 4Subagents retry transient failures locally with a bounded number of attempts and propagate only what they cannot resolve, together with what they tried and what they already found.
- 5The synthesis output carries coverage annotations: which findings are well-supported and which topic areas have gaps because sources were unavailable.
What the exam guide checks here
Objective wording from the exam guide (English).
- 5.3.K1Structured error context (type, query, partial results, alternatives) enables coordinator recovery
- 5.3.K2Access failures vs valid empty results
- 5.3.K3Generic statuses hide context from the coordinator
- 5.3.K4Silent suppression and killing the whole workflow are both anti-patterns
- 5.3.S1Return failure type, attempt, partial results, alternatives
- 5.3.S2Report access failures distinctly from empty results
- 5.3.S3Local recovery for transient failures, propagate only the unresolved
- 5.3.S4Synthesis output annotates coverage: well-supported findings vs gaps
Lock it in
Answer scenario questions on this topic. Every option comes with an explanation.
Practise 5.3 · 21 questions